Proteins self-assemble into crystals, gels, amyloid fibrils, amorphous aggregates, dense liquid droplets with implications in several fields such as biotechnology, condensation diseases and the food industry. Self-assembly processes are dictated by protein self-interactions; however, proteins are anisotropic particles, i.e. their surface is chemically heterogeneous. Hence, protein-protein interactions are strongly orientation-dependent and protein self-assembly is dramatically influenced by surface modifications. We examine this topic firstly using a model protein, Human γD-crystallin (HGD), and then on a novel type of Virus-Like Particle (VLP), ADDomer, a promising vaccine candidate. Human γD-crystallin (HGD) is an eye-lens protein and along with other crystallins, it creates the refractive index gradient necessary for proper lens function. Naturally occurring surface modifications of HGD trigger aggregation/crystallization, leading to age-related or hereditary cataract onset. For these reasons, the influence of surface modifications (such as single-point mutations) on HGD self-assembly have been extensively studied making this protein a perfect model to study protein surface anisotropy and its role in directing protein assembly. Cys-110 is the only surface exposed cysteine of HGD and hence responsible for covalent dimerization, a process that contributes to age-related cataract onset. However, the biological advantage of Cys-110 is not clear. Therefore, we mutated Cys to Met and Ser (C110M and C110S) and studied the mutant protein self-assembly. We found that C110S is recalcitrant to crystallization, while the mutant C110M crystallizes promptly; C110M crystals have the same characteristics as those of HGD, however, we suggest that the different nucleation behaviour of C110M may be due to subtle changes in the water shell of 110th site and its hydrophobicity. In conclusion, Cys-110 has the ability to suppress HGD crystallization; we speculate that the presence of Cys-110 is advantageous compared to either Met or Ser since Met enhances crystallization, hindering protein long-term stability, while Ser would decrease the HGD refractive index increment, which is essential for eye-lens function. The effects of other single-point mutations on HGD self-assembly are known, for example R58H and R36S enhance crystallization, mutations at the 23rd site (e.g. P23V, P23T) induce the formation of retrograde solubility assemblies (i.e. they melt when cooled). It has been shown that the self-assembly behaviour of the double mutants P23VR36S, P23TR58H and P23VR58H can be predicted by those of the respective single mutants. To generalize this, we studied a novel double mutant, P23VC110M, that both enhances crystallization of the protein due to the C110M mutation and at higher temperatures forms reversible assemblies with retrograde solubility due to the presence of the P23Vmutation. P23VC110M forms both these phases and a new polymorph, needle/plate-shaped crystals, which possibly arise from new crystal contacts involving the two mutation sites. Hence, we confirmed that the self-assembly of double mutants can be predicted from the behaviour of the respective single mutants, but also more complex scenarios can arise. We also examined the morphology of the retrograde solubility assemblies formed due to mutation P23V, which unusually are perfectly spherical. Using P23VC110M, we compared the properties of the amorphous, large, reversible and spherical assemblies to other forms of protein spherical superstructures, i.e. particulates and amyloid spherulites. P23VC110M spherical assemblies are not formed by amyloid fibrils but are amorphous and cannot be ascribed to any of the protein superstructures already known in literature. We therefore suggest that this is a new class of protein assembly. The effect of surface modifications on protein self-assembly was then assessed for a novel type of Virus-like particle (VLP), ADDomer, based on Human Adenovirus serotype-3 (HAdv-3) penton-base protein. VLPs can be modified to display on their surface multiple copies of an epitope and hence trigger immune-response against a specific disease. However, vaccines often need to be stored in extremely cold conditions, which limits their distribution in remote areas of the world. We have studied the effects of temperature on different variations of ADDomer VLPs to understand the drivers for their self-assembly and their thermostabiliy. The thermal stability of ADDomer was compared to the one of a similar VLP, ChADDomer, derived from Chimpanzee Adenovirus serotype-3 (ChAdV-3); mutants L56C and S57C of ChADDomer were also designed to promote the formation of inter-penton disulphide bridges that could hinder VLP disassembly. A hybrid construct between ADDomer and ChADDomer, Chimera was also tested, along with its disulphide bridge-forming mutant S57C. We found that, of all the VLPs studied, ChADDomer S57C and L56C are the most thermally stable against aggregation and that disulphide-bridge promoting mutations also help stabilizing protein secondary structure. Conversely, ADDomer and Chimera abruptly aggregate. By structural comparison with ChADDomer, we identified the protein regions that may be responsible for this behaviour.